Bullpen defines 3 tools with complete JSON schemas and basic descriptions. However, descriptions are minimal (10-60 chars), parameter descriptions are entirely absent, and error handling guidance is not visible. Tool names follow verb_noun convention (submit_*), but parameter documentation is sparse. No evidence of output schema documentation, pagination support, or recovery guidance. Schemas are present but lack parameter-level descriptions critical for LLM reasoning.

• Add 150-250 character descriptions to each tool explaining WHAT it does, WHEN to call it, and WHAT it returns. Example: 'submit_research_plan: Submit your interpreted research plan before analysis. Call this after deciding on research intent, decision criteria, and planned checks. Returns confirmation and plan ID for downstream analysis steps.'
• Add descriptions to every parameter. For 'intent' enum, document each value: 'intent (enum): Research scope, single_equity (one stock), compare_equities (2+ stocks), sector_analysis (industry), macro_theme (economic theme), watchlist (tracked list), portfolio (holdings), general_research (open-ended).'
• Document output schemas for each tool. Specify what fields are returned (e.g., plan_id, status, validation_errors) and their types so agents can chain results.
• Add error handling guidance: 'If submission fails due to invalid criteria, the tool returns error_code and error_details. Retry with refined criteria or call search_research_templates() to see valid patterns.'
• Add parameter constraints: mark 'confidence' as 0.0-1.0 range, 'minItems' for arrays, and document required vs optional fields. Use descriptions to explain why each constraint exists.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry